The Hidden Cost of AI: Unmasking its Environmental and Social Footprint

The exponential growth of AI, particularly large language models, is driving a profound environmental crisis and exacerbating social injustices.

Article link: https://open.substack.com/pub/johanosteyn/p/the-hidden-cost-of-ai-unmasking-its?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Audio summary: https://youtu.be/uyJSF9lhqvM

I write about various issues of interest to me that I want to bring to the reader’s attention. While my main work is in Artificial Intelligence and technology, I also cover areas around politics, education, and the future of our children. This article delves into the critical, often overlooked, environmental and social impact of large language models, a profound concern for the sustainability of our planet and the well-being of vulnerable communities, directly impacting the future our children will inherit.

The exponential acceleration of artificial intelligence—particularly large language models (LLMs) and generative AI—has unleashed a profound environmental crisis centred on the massive data centres required to train, deploy, and operate these systems. These facilities consume extraordinary amounts of electricity and water. Recent projections indicate that AI-specific energy demand could reach 165–326 terawatt-hours per year by 2028, equivalent to powering 22% of all U.S. households.

Training a single model such as GPT-3 required 1,287 MWh of electricity, emitted 552 tonnes of CO₂, and evaporated over 700,000 litres of freshwater. This is not merely a technical challenge; it is a defining ethical, social, and political issue that demands our immediate attention.

CONTEXT AND BACKGROUND

The rapid advancement and widespread adoption of generative AI models have brought undeniable benefits, from improving worker productivity to accelerating scientific research. However, this “AI gold rush” comes with significant environmental consequences that are often difficult to quantify and mitigate. Data centres, the physical infrastructure housing AI servers, are now consuming electricity at unprecedented rates.

Globally, electricity consumption from data centres is estimated to be around 415 terawatt-hours (TWh) in 2024, representing about 1.5% of global electricity consumption, and is projected to double to around 945 TWh by 2030. This surge is largely driven by AI, which could account for 35-50% of data centre power use by 2030. In the US alone, data centres consumed 183 TWh in 2024, over 4% of the country’s total, projected to grow by 133% by 2030.

The footprint of everyday AI use is equally staggering. ChatGPT alone uses nearly 148 million litres of water daily, with each 100-word response consuming approximately 519 ml of water and 0.14 kWh of electricity. This rapid surge in demand is forcing utilities to delay the retirement of coal-fired power plants and build over 200 new fossil-fuel generators, increasing global carbon emissions.

Beyond energy and water, the manufacturing of AI chips requires vast amounts of critical minerals and generates significant e-waste, with one study projecting 16 million tons of cumulative e-waste by 2030 from generative AI alone.

INSIGHT AND ANALYSIS

Most troubling is the geographical distribution of these burgeoning data centres. Across the United States—and increasingly globally—data centres are disproportionately located in poor Black and brown communities, creating new forms of environmental injustice. Over $200 billion in infrastructure is being placed in regions historically treated as “sacrifice zones,” where residents already face water scarcity, polluted air, and weak regulatory protections.

Communities near these data centres report brown tap water, dry wells, rising utility costs, diesel generator pollution, and increased respiratory illness due to heightened particulate matter and nitrogen dioxide. More than 160 new AI data centres have been constructed in water-stressed areas globally, a 70% increase, with similar patterns observed in Uruguay, Chile, Mexico, and drought-stricken parts of the U.S..Despite promises of economic uplift and job creation, these communities overwhelmingly experience environmental degradation while tech giants claim substantial tax breaks and extract local resources at an industrial scale.

This directly impacts the future of our country, as the pursuit of AI without equitable development can deepen existing inequalities, particularly in South Africa where resource scarcity and social justice are pressing concerns.

For me personally, as a father, the thought of our children inheriting a world where technological advancement comes at the cost of environmental integrity and social equity is deeply unsettling. The lack of transparency from major AI providers regarding their energy and water consumption further exacerbates this problem, making it difficult to accurately assess AI’s true carbon footprint.

IMPLICATIONS

Addressing the hidden costs of AI requires a multi-faceted and urgent global response. Firstly, governments must develop robust regulatory frameworks that mandate transparency from AI companies regarding their direct environmental consequences, including energy, water, and e-waste across the entire lifecycle of AI models and hardware. The EU AI Act, while a step, has been criticised for diluting environmental provisions, highlighting the need for stronger, explicit regulations.

Secondly, there must be a concerted effort to shift data centres to renewable energy sources and implement advanced water-efficient cooling technologies, such as closed-loop systems and the use of non-potable water.

Thirdly, the ethical siting of data centres must become a priority, ensuring that vulnerable communities are not disproportionately burdened by environmental costs. This requires engaging directly with affected communities and incorporating their concerns into policy decisions.

Finally, while AI poses environmental risks, it also offers solutions. AI can be harnessed for climate modelling, optimising energy grids, detecting deforestation, and improving sustainable agriculture. This dual nature means we must pursue “Green AI” solutions, making algorithms more efficient and promoting sustainable applications. This proactive approach is essential for the future of our country, ensuring that AI contributes meaningfully to sustainable development and safeguards the well-being of our children.CLOSING

TAKEAWAY

The environmental and social footprint of AI is a critical, growing concern. We must demand transparency, implement robust regulations, and strategically harness AI’s potential for sustainable solutions to ensure a just and equitable digital future for all.

Author Bio: Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net